14 KiB
14 KiB
from __future__ import annotations
from dataclasses import dataclass
from typing import Any, Dict, List, Optional, Tuple
import numpy as np
import pandas as pd
from statsmodels.tsa.stattools import adfuller, coint
from statsmodels.tsa.vector_ar.vecm import coint_johansen
from statsmodels.tsa.vector_ar.vecm import coint_johansen # type: ignore
# ---
from cvttpy_tools.base import NamedObject
from cvttpy_tools.config import Config
from cvttpy_tools.logger import Log
from cvttpy_tools.timeutils import NanoPerSec, SecPerHour, current_nanoseconds
from cvttpy_tools.web.rest_client import RESTSender
# ---
from cvttpy_trading.trading.instrument import ExchangeInstrument
from cvttpy_trading.trading.mkt_data.md_summary import MdTradesAggregate, MdSummary
@dataclass
class InstrumentQuality(NamedObject):
instrument_: ExchangeInstrument
record_count_: int
latest_tstamp_: Optional[pd.Timestamp]
status_: str
reason_: str
@dataclass
class PairStats(NamedObject):
instrument_a_: ExchangeInstrument
instrument_b_: ExchangeInstrument
pvalue_eg_: Optional[float]
pvalue_adf_: Optional[float]
pvalue_j_: Optional[float]
trace_stat_j_: Optional[float]
rank_eg_: int = 0
rank_adf_: int = 0
rank_j_: int = 0
composite_rank_: int = 0
def as_dict(self) -> Dict[str, Any]:
return {
"instrument_a": self.instrument_a_.instrument_id(),
"instrument_b": self.instrument_b_.instrument_id(),
"pvalue_eg": self.pvalue_eg_,
"pvalue_adf": self.pvalue_adf_,
"pvalue_j": self.pvalue_j_,
"trace_stat_j": self.trace_stat_j_,
"rank_eg": self.rank_eg_,
"rank_adf": self.rank_adf_,
"rank_j": self.rank_j_,
"composite_rank": self.composite_rank_,
}
class DataFetcher(NamedObject):
sender_: RESTSender
interval_sec_: int
history_depth_sec_: int
def __init__(
self,
base_url: str,
interval_sec: int,
history_depth_sec: int,
) -> None:
self.sender_ = RESTSender(base_url=base_url)
self.interval_sec_ = interval_sec
self.history_depth_sec_ = history_depth_sec
def fetch(self, exch_acct: str, inst: ExchangeInstrument) -> List[MdTradesAggregate]:
rqst_data = {
"exch_acct": exch_acct,
"instrument_id": inst.instrument_id(),
"interval_sec": self.interval_sec_,
"history_depth_sec": self.history_depth_sec_,
}
response = self.sender_.send_post(endpoint="md_summary", post_body=rqst_data)
if response.status_code not in (200, 201):
Log.error(
f"{self.fname()}: error {response.status_code} for {inst.details_short()}: {response.text}")
return []
mdsums: List[MdSummary] = MdSummary.from_REST_response(response=response)
return [
mdsum.create_md_trades_aggregate(
exch_acct=exch_acct, exch_inst=inst, interval_sec=self.interval_sec_
)
for mdsum in mdsums
]
class QualityChecker(NamedObject):
interval_sec_: int
def __init__(self, interval_sec: int) -> None:
self.interval_sec_ = interval_sec
def evaluate(self, inst: ExchangeInstrument, aggr: List[MdTradesAggregate]) -> InstrumentQuality:
if len(aggr) == 0:
return InstrumentQuality(
instrument_=inst,
record_count_=0,
latest_tstamp_=None,
status_="FAIL",
reason_="no records",
)
aggr_sorted = sorted(aggr, key=lambda a: a.aggr_time_ns_)
latest_ts = pd.to_datetime(aggr_sorted[-1].aggr_time_ns_, unit="ns", utc=True)
now_ts = pd.Timestamp.utcnow()
recency_cutoff = now_ts - pd.Timedelta(seconds=2 * self.interval_sec_)
if latest_ts <= recency_cutoff:
return InstrumentQuality(
instrument_=inst,
record_count_=len(aggr_sorted),
latest_tstamp_=latest_ts,
status_="FAIL",
reason_=f"stale: latest {latest_ts} <= cutoff {recency_cutoff}",
)
gaps_ok, reason = self._check_gaps(aggr_sorted)
status = "PASS" if gaps_ok else "FAIL"
return InstrumentQuality(
instrument_=inst,
record_count_=len(aggr_sorted),
latest_tstamp_=latest_ts,
status_=status,
reason_=reason,
)
def _check_gaps(self, aggr: List[MdTradesAggregate]) -> Tuple[bool, str]:
NUM_TRADES_THRESHOLD = 50
if len(aggr) < 2:
return True, "ok"
interval_ns = self.interval_sec_ * NanoPerSec
for idx in range(1, len(aggr)):
prev = aggr[idx - 1]
curr = aggr[idx]
delta = curr.aggr_time_ns_ - prev.aggr_time_ns_
missing_intervals = int(delta // interval_ns) - 1
if missing_intervals <= 0:
continue
prev_nt = prev.num_trades_
next_nt = curr.num_trades_
estimate = self._approximate_num_trades(prev_nt, next_nt)
if estimate > NUM_TRADES_THRESHOLD:
return False, (
f"gap of {missing_intervals} interval(s), est num_trades={estimate} > {NUM_TRADES_THRESHOLD}"
)
return True, "ok"
@staticmethod
def _approximate_num_trades(prev_nt: int, next_nt: int) -> float:
if prev_nt is None and next_nt is None:
return 0.0
if prev_nt is None:
return float(next_nt)
if next_nt is None:
return float(prev_nt)
return (prev_nt + next_nt) / 2.0
class PairAnalyzer(NamedObject):
price_field_: str
interval_sec_: int
def __init__(self, price_field: str, interval_sec: int) -> None:
self.price_field_ = price_field
self.interval_sec_ = interval_sec
def analyze(self, series: Dict[ExchangeInstrument, pd.DataFrame]) -> List[PairStats]:
instruments = list(series.keys())
results: List[PairStats] = []
for i in range(len(instruments)):
for j in range(i + 1, len(instruments)):
inst_a = instruments[i]
inst_b = instruments[j]
df_a = series[inst_a][["tstamp", "price"]].rename(
columns={"price": "price_a"}
)
df_b = series[inst_b][["tstamp", "price"]].rename(
columns={"price": "price_b"}
)
merged = pd.merge(df_a, df_b, on="tstamp", how="inner").sort_values(
"tstamp"
)
stats = self._compute_stats(inst_a, inst_b, merged)
if stats:
results.append(stats)
self._rank(results)
return results
def _compute_stats(
self,
inst_a: ExchangeInstrument,
inst_b: ExchangeInstrument,
merged: pd.DataFrame,
) -> Optional[PairStats]:
if len(merged) < 2:
return None
px_a = merged["price_a"].astype(float)
px_b = merged["price_b"].astype(float)
std_a = float(px_a.std())
std_b = float(px_b.std())
if std_a == 0 or std_b == 0:
return None
z_a = (px_a - float(px_a.mean())) / std_a
z_b = (px_b - float(px_b.mean())) / std_b
p_eg: Optional[float]
p_adf: Optional[float]
p_j: Optional[float]
trace_stat: Optional[float]
try:
p_eg = float(coint(z_a, z_b)[1])
except Exception as exc:
Log.warning(f"{self.fname()}: EG failed for {inst_a.details_short()}/{inst_b.details_short()}: {exc}")
p_eg = None
try:
spread = z_a - z_b
p_adf = float(adfuller(spread, maxlag=1, regression="c")[1])
except Exception as exc:
Log.warning(f"{self.fname()}: ADF failed for {inst_a.details_short()}/{inst_b.details_short()}: {exc}")
p_adf = None
try:
data = np.column_stack([z_a, z_b])
res = coint_johansen(data, det_order=0, k_ar_diff=1)
trace_stat = float(res.lr1[0])
cv10, cv5, cv1 = res.cvt[0]
if trace_stat > cv1:
p_j = 0.01
elif trace_stat > cv5:
p_j = 0.05
elif trace_stat > cv10:
p_j = 0.10
else:
p_j = 1.0
except Exception as exc:
Log.warning(f"{self.fname()}: Johansen failed for {inst_a.details_short()}/{inst_b.details_short()}: {exc}")
p_j = None
trace_stat = None
return PairStats(
instrument_a_=inst_a,
instrument_b_=inst_b,
pvalue_eg_=p_eg,
pvalue_adf_=p_adf,
pvalue_j_=p_j,
trace_stat_j_=trace_stat,
)
def _rank(self, results: List[PairStats]) -> None:
self._assign_ranks(results, key=lambda r: r.pvalue_eg_, attr="rank_eg_")
self._assign_ranks(results, key=lambda r: r.pvalue_adf_, attr="rank_adf_")
self._assign_ranks(results, key=lambda r: r.pvalue_j_, attr="rank_j_")
for res in results:
res.composite_rank_ = res.rank_eg_ + res.rank_adf_ + res.rank_j_
results.sort(key=lambda r: r.composite_rank_)
@staticmethod
def _assign_ranks(
results: List[PairStats], key, attr: str
) -> None:
values = [key(r) for r in results]
sorted_vals = sorted([v for v in values if v is not None])
for res in results:
val = key(res)
if val is None:
setattr(res, attr, len(sorted_vals) + 1)
continue
rank = 1 + sum(1 for v in sorted_vals if v < val)
setattr(res, attr, rank)
class PairSelectionEngine(NamedObject):
config_: object
instruments_: List[ExchangeInstrument]
price_field_: str
fetcher_: DataFetcher
quality_: QualityChecker
analyzer_: PairAnalyzer
interval_sec_: int
history_depth_sec_: int
data_quality_cache_: List[InstrumentQuality]
pair_results_cache_: List[PairStats]
def __init__(
self,
config: Config,
instruments: List[ExchangeInstrument],
price_field: str,
) -> None:
self.config_ = config
self.instruments_ = instruments
self.price_field_ = price_field
interval_sec = int(config.get_value("interval_sec", 0))
history_depth_sec = int(config.get_value("history_depth_hours", 0)) * SecPerHour
base_url = config.get_value("cvtt_base_url", None)
assert interval_sec > 0, "interval_sec must be > 0"
assert history_depth_sec > 0, "history_depth_sec must be > 0"
assert base_url, "cvtt_base_url must be set"
self.fetcher_ = DataFetcher(
base_url=base_url,
interval_sec=interval_sec,
history_depth_sec=history_depth_sec,
)
self.quality_ = QualityChecker(interval_sec=interval_sec)
self.analyzer_ = PairAnalyzer(price_field=price_field, interval_sec=interval_sec)
self.interval_sec_ = interval_sec
self.history_depth_sec_ = history_depth_sec
self.data_quality_cache_ = []
self.pair_results_cache_ = []
async def run_once(self) -> None:
quality_results: List[InstrumentQuality] = []
price_series: Dict[ExchangeInstrument, pd.DataFrame] = {}
for inst in self.instruments_:
exch_acct = inst.user_data_.get("exch_acct") or inst.exchange_id_
aggr = self.fetcher_.fetch(exch_acct=exch_acct, inst=inst)
q = self.quality_.evaluate(inst, aggr)
quality_results.append(q)
if q.status_ != "PASS":
continue
df = self._to_dataframe(aggr, inst)
if len(df) > 0:
price_series[inst] = df
self.data_quality_cache_ = quality_results
self.pair_results_cache_ = self.analyzer_.analyze(price_series)
def _to_dataframe(self, aggr: List[MdTradesAggregate], inst: ExchangeInstrument) -> pd.DataFrame:
rows: List[Dict[str, Any]] = []
for item in aggr:
rows.append(
{
"tstamp": pd.to_datetime(item.aggr_time_ns_, unit="ns", utc=True),
"price": self._extract_price(item, inst),
"num_trades": item.num_trades_,
}
)
df = pd.DataFrame(rows)
return df.sort_values("tstamp").reset_index(drop=True)
def _extract_price(self, aggr: MdTradesAggregate, inst: ExchangeInstrument) -> float:
price_field = self.price_field_
# MdTradesAggregate inherits hist bar with fields open_, high_, low_, close_, vwap_
field_map = {
"open": aggr.open_,
"high": aggr.high_,
"low": aggr.low_,
"close": aggr.close_,
"vwap": aggr.vwap_,
}
raw = field_map.get(price_field, aggr.close_)
return inst.get_price(raw)
def sleep_seconds_until_next_cycle(self) -> float:
now_ns = current_nanoseconds()
interval_ns = self.interval_sec_ * NanoPerSec
next_boundary = (now_ns // interval_ns + 1) * interval_ns
return max(0.0, (next_boundary - now_ns) / NanoPerSec)
def quality_dicts(self) -> List[Dict[str, Any]]:
res: List[Dict[str, Any]] = []
for q in self.data_quality_cache_:
res.append(
{
"instrument": q.instrument_.instrument_id(),
"record_count": q.record_count_,
"latest_tstamp": q.latest_tstamp_.isoformat() if q.latest_tstamp_ else None,
"status": q.status_,
"reason": q.reason_,
}
)
return res
def pair_dicts(self) -> List[Dict[str, Any]]:
return [p.as_dict() for p in self.pair_results_cache_]